Look, we just need to add some new 'planes' to Unicode - that mirror all communicatively-useful characters, but with extra state bits for...
guaranteed human output - anyone who emits text in these ranges that was AI generated, rather than artisanally human-composed, goes straight to jail.
for human eyes only - anyone who lets any AI train on, or even consider, any text in these ranges goes straight to jail. Fnord, "that doesn't look like anything to me".
admittedly AI generated - all AI output must use these ranges as disclosure, or – you guessed it - those pretending otherwise go straight to jail.
Of course, all the ranges generate visually-indistinguishable homoglyphs, so it's a strictly-software-mediated quasi-covert channel for fair disclosure.
When you cut & paste text from various sources, the provenance comes with it via the subtle character encoding differences.
I am only (1 - epsilon) joking.
guaranteed human output - anyone who emits text in these ranges that was AI generated, rather than artisanally human-composed, goes straight to jail.
for human eyes only - anyone who lets any AI train on, or even consider, any text in these ranges goes straight to jail. Fnord, "that doesn't look like anything to me".
admittedly AI generated - all AI output must use these ranges as disclosure, or – you guessed it - those pretending otherwise go straight to jail.
Of course, all the ranges generate visually-indistinguishable homoglyphs, so it's a strictly-software-mediated quasi-covert channel for fair disclosure.
When you cut & paste text from various sources, the provenance comes with it via the subtle character encoding differences.
I am only (1 - epsilon) joking.